Climate Change and Health Seminar: Tools for environmentally-informed malaria control in the Western Amazon

Transcript

• 00:00<v ->Which is hosted by the Yale Center on Continent Health.</v>
• 00:04So today we have a hybrid seminar due to the COVID pandemic,
• 00:11so we have the students joining us in person,
• 00:16but also for the students who could not join us,
• 00:19they can also join us online (indistinct).
• 00:26But before we move on,
• 00:28I just want to have two quick kind of housekeeping rules.
• 00:33So, you guys have submitted questions to our speakers.
• 00:36So at the end, we will have a Q&amp;A session,
• 00:39so you guys feel free to ask your question,
• 00:41raise your hand so the speaker
• 00:43can actually hear you quite clearly.
• 00:46And for the folks online, if you have any questions,
• 00:50also please don't hesitate put them in the chat box.
• 00:55And we will also go through those questions
• 00:57on behalf of the attendants.
• 01:00So, it's my great pleasure today to introduce
• 01:05our first speaker of the seminar series,
• 01:08Dr. Benjamin Zaitchik.
• 01:11Dr. Zaitchik is a Professor in the Department of Earth
• 01:14and Planetary Sciences at the Johns Hopkins University.
• 01:18His research addresses hydro-climatic variety,
• 01:22including fundamental work on atmospheric science
• 01:25and hydrological processes,
• 01:27and application to program on water resources,
• 01:30agriculture and human health.
• 01:34Dr. Zaitchik is actually also the President
• 01:38of the Two House Session
• 01:40of the American Geophysical Union, in short AGU.
• 01:45So another thing he want to mention
• 01:48is Ben actually got his PhD from here in 2006,
• 01:53from the Department of Geology and Geophysics.
• 01:58So we are very pleased to welcome back Ben
• 02:03at Yale, although virtually.
• 02:05So without further ado, let's welcome Dr. Benjamin Zaitchik.
• 02:12<v Benjamin>Great. Thanks so much Kai.</v>
• 02:13And thank you for the opportunity to speak.
• 02:16You know, I have to admit,
• 02:18I've somewhat enjoyed this remote world
• 02:20and our ability to talk and interact at a distance,
• 02:22but I was a little disappointed when I'm not able
• 02:25to be up there in Newhaven right now,
• 02:28because it would've been fun to come back.
• 02:29As Kai mentioned, I did do my PhD there,
• 02:32but not in public health, kind of cross 34 on Science Hill
• 02:35in geology and geophysics.
• 02:37But while I was there,
• 02:39I was not yet working in the geo health area,
• 02:41but I got to see a lot of collaboration going on,
• 02:44particularly between Durland Fish
• 02:46and some of his students in public health,
• 02:48and my geology department,
• 02:50which really was my first exposure to this idea
• 02:52that you could really make use of some
• 02:54of our environmental information analyses
• 02:57to inform infectious disease analysis.
• 03:02So the talk today,
• 03:03I'm going to be focusing on malaria in the Western Amazon.
• 03:07I've got a long list of names here,
• 03:09that's only a partial list.
• 03:11I want to particularly acknowledge Bill Pan
• 03:13at Duke University,
• 03:14who has led most of the work I'm going to present on today
• 03:17from the epidemiological side, as well as Mark Janko,
• 03:21Cristina Recalde, and Francisco Pizzitutti,
• 03:23whose results I will be showing.
• 03:28So, might start with some deep background
• 03:32and perhaps an apology in that the idea
• 03:35that malaria somehow in an environmentally mediated disease
• 03:41is not particularly new, right?
• 03:42It shouldn't come as a surprise to anybody.
• 03:46In ancient times,
• 03:47malaria was associated with the rise of Sirius,
• 03:51the dog star,
• 03:53which would come in the days of mid to late summer,
• 03:56around the Mediterranean, where the Greeks and others
• 03:59were studying this and aware of its impact.
• 04:02That is why we call them the dog days of summer,
• 04:04because that's when Sirius became visible.
• 04:06And you can see writings about this across Mediterranean
• 04:11at the time.
• 04:12Hippocrates, who was famously very interested
• 04:14in the relationship between environment and meteorology
• 04:18and health wrote specifically about how
• 04:20these cyclical fevers that we now understand
• 04:23to be malaria were associated with the season,
• 04:26and clearly understood quite clearly
• 04:28that this was not an astrological phenomenon,
• 04:30but that this was a phenomenon tied to the seasonality
• 04:33and to the oppressive heat built at that time.
• 04:37Now, this is millennia before the mosquito-mediated
• 04:40pathway of malaria transmission was confirmed,
• 04:45as well as before the plasmodium was identified,
• 04:48certainly as the parasite.
• 04:50And yet this understanding that malaria
• 04:52was sensitive to these changes was clear.
• 04:56I mean, the very fact we call it malaria, right? Bad air.
• 04:59It's the disease that is most associated inherently
• 05:02in our naming system with this idea
• 05:04of environmental sensitivity.
• 05:07And so, you might think that we had this
• 05:08kind of figured out, right?
• 05:09So why in the year 2021,
• 05:11am I here to talk to you about our attempts
• 05:14and our struggles to continue to understand
• 05:16in a predictive fashion,
• 05:18the way in which malaria responds
• 05:20to environmental variability?
• 05:22And I think the answer is that it's a bit complicated.
• 05:24And so what I'm going to talk about here
• 05:26is something where we really need to understand
• 05:29the environmental influence,
• 05:31and the climatic influence as well as
• 05:32other environmental influences,
• 05:34in the full context of a coupled natural human system
• 05:37that evolves with time.
• 05:39And so, simply understand that malaria has the potential
• 05:41to be sensitive to environmental factors
• 05:44is not in and of itself a useful or actionable
• 05:47predictive system.
• 05:49So the talk today, I'm going to start off
• 05:51with some background on malaria in the Western Amazon,
• 05:54and apology in advance if doing so is insulting
• 05:58to folks in public health who have a deep understanding
• 06:00of malaria in this region,
• 06:01but I'm not sure of everyone's background.
• 06:03So we'll go through a little bit of that history
• 06:05and current dynamics.
• 06:07Then, I'm going to spend a little bit more time
• 06:08than you probably want me to on physical geography
• 06:11and hydrometeorology because that's really
• 06:13what I bring to these set of analyses.
• 06:18Then I'll move on and just give three of the cases
• 06:21in which you've tried to integrate
• 06:23these kinds of environmental information systems
• 06:25to our understanding and forecast of malaria in this region.
• 06:30And I want to emphasize something that Kai said,
• 06:32was certainly type into the chat
• 06:34if you would like to say anything.
• 06:35Also feel free just to unmute and interrupt
• 06:38if I say something that is unclear.
• 06:44So again, based back on the malaria in the Amazon,
• 06:46this is from the malaria Atlas.
• 06:49And what we see here is that the dominant type of malaria
• 06:52will be vivax that is present
• 06:55throughout the Amazon basin,
• 06:57but you also see falciparum in some concentration,
• 07:01and the Western Amazon part of Peru and Western Brazil,
• 07:05and then we focusing on, you will see both
• 07:09in significant amounts.
• 07:13I should note that I'm zoomed in here on the Amazon basin.
• 07:17The Amazon is home to over 90% of malaria
• 07:20in the Western hemisphere.
• 07:21And so it's really in terms of studying the Americas,
• 07:26it's the place that one would want to be focusing
• 07:29a lot of effort on malaria reduction.
• 07:33And in this region,
• 07:36malaria is classically associated with deforestation,
• 07:40encroachment into the natural forest.
• 07:42So, it's just a satellite time-lapse,
• 07:44showing over about 30 years
• 07:47what we all know to be true, this massive deforestation.
• 07:49This particular lapse rate is from Brazil.
• 07:51You see similar things throughout the Amazon basin.
• 07:54Classic pattern here is a road gets built,
• 07:56you surmise from that flash across the screen
• 07:58at the beginning of the time series.
• 07:59Once the road is built, you get this herringbone pattern
• 08:02of deforestation as land is cleared for logging,
• 08:05but then also for agriculture and ranching.
• 08:08And this dynamic was associated
• 08:11with a massive burst of malaria in the Amazon region,
• 08:15particularly in the '80s and 1990s.
• 08:18And so that was really the time
• 08:19of the most rapid deforestation going on
• 08:21over much of the Amazon.
• 08:23Continues to be a major issue today,
• 08:25but that's when the rate was the highest.
• 08:28And what you had there was a situation
• 08:30where epidemiologically naive populations
• 08:33were entering into a region where the anopheles mosquitoes,
• 08:38the dominant vector of malaria,
• 08:42were present in large numbers,
• 08:44and the kinds of livelihoods we were seeing in particular,
• 08:46this kind of entering the wilderness for logging and such,
• 08:50and then a lot of mobility going on
• 08:52all led to this really strong epidemic peak.
• 08:56And from observing the dynamics,
• 09:00this what now we would consider
• 09:02to be a classic hypothesis emerged
• 09:04called the malaria frontier.
• 09:06And so you have frontier malaria in situations
• 09:09where you have populations that do not have immunity,
• 09:13and who do not have behavioral patterns
• 09:15associated with trying to avoid malaria,
• 09:16because they're new to the area,
• 09:18enter into the wilderness frontier,
• 09:21and you get this burst of epidemic peaks,
• 09:25followed by a gradual adjustment
• 09:28as you get some resistance building up,
• 09:30as you get populations' behavior changing,
• 09:32and as you get livelihood changes
• 09:35that maybe are a little less mobile
• 09:36and include less interface with wildlands,
• 09:40and you settle into an endemic pattern,
• 09:42this endemic malaria.
• 09:44And so, you know, this has flashed through enough times
• 09:46that maybe you've noticed by now
• 09:47that you can kind of see the timing, right?
• 09:49While things change throughout the time series I'm showing,
• 09:52after about the year 2000 or so,
• 09:55the change isn't as rampant.
• 09:57You don't see as much clear cutting, right?
• 09:58That mostly happened in the '80s and '90s.
• 10:01Again, this is a time series from Brazil.
• 10:03You'd see similar things in the parts of Peru and Ecuador
• 10:06that we're focusing on.
• 10:08So, when I talk about malaria today,
• 10:11I'm going to be focusing on the last 20 years,
• 10:12which is really coast frontier malaria.
• 10:15Okay, so this is the time where we say, okay,
• 10:17we've kind of been through that initial burst of malaria
• 10:21that happens when you enter the frontier.
• 10:24And now, we're in the situation where we are looking
• 10:26at transmission patterns in populations
• 10:28that I wouldn't say that it's a stable population.
• 10:32There's always movement going on.
• 10:33But you're no longer talking about this encroachment.
• 10:35You're talking about interfaces within
• 10:38what is more or less a settled area.
• 10:44Okay, and so what does that look like
• 10:46if you just look at case numbers in the Amazon?
• 10:48So here, I'm showing a time series from 2000 on.
• 10:51And so what you're listing over to the left here
• 10:53are there really high numbers that preceded this?
• 10:55So the numbers on this curve, you can kind of see Brazil,
• 10:58that red curve coming down, right,
• 11:00from what was a really big peak in the 1990s.
• 11:03And if you ignore Venezuela,
• 11:05which as we all know has had its own challenges,
• 11:08you would generally say,
• 11:09"Oh, this is kind of a story of cases falling, okay,
• 11:12from that frontier malaria peak."
• 11:16But if you look a little more closely,
• 11:19over the last 20 years, you'll see that progress
• 11:21has stalled and even reversed.
• 11:23And so expanding the Y axes a little bit here
• 11:26to look at Columbia, Ecuador and Peru,
• 11:28just over the past 15 years or so,
• 11:32what you see is a rather significant peak in Ecuador.
• 11:35It came down a little bit after that
• 11:36but it's come back up.
• 11:38Peru, quite a significant percent wise increase,
• 11:40because the case has got so low in the the early 2010s.
• 11:48Sorry, that was Ecuador.
• 11:49Big, significant increase in Ecuador.
• 11:51I missed my labels here.
• 11:53Then bottom one is Peru showing
• 11:55the significant increase, again.
• 11:56And so you see these large percent wise increase
• 11:58in these Western Amazonian countries.
• 12:02Focusing on Peru specifically for a moment,
• 12:05because that's what a bunch of our data
• 12:07are going to come from, that I'm going to show
• 12:08in the next section.
• 12:10What you see here is a phenomenon where, again,
• 12:13cases were quite high in the 1990s,
• 12:15but there seemed to be a period where you were at
• 12:17a kind of a stable level in the 2000s,
• 12:20and then a rapid decline to the point where
• 12:22it was really getting close to elimination around 2010,
• 12:25before it burst back up.
• 12:26And so now what's been happening?
• 12:28So that period, as I'll get to it towards
• 12:31the end of the talk,
• 12:33was a period of a significant intervention
• 12:35and attempt to eliminate malaria from this region.
• 12:39So the PAMAFRO program,
• 12:40which ran for about five years involved
• 12:43a number of malaria control activities.
• 12:46Again, details come later, and it really did seem to work.
• 12:50Then in 2011, you had this historical flood.
• 12:53There was a flood of record over much of the Amazon,
• 12:55the biggest one in the observed record.
• 12:58And it had tremendous impacts across the region.
• 13:02But one thing that happened was what we saw
• 13:04an increase in malaria cases, this reversal, okay?
• 13:09Now this flood coincided with the end
• 13:10of the PAMAFRO program.
• 13:12And so we have some disentangling to do,
• 13:14about what's going on when it increased.
• 13:16And when this first happened,
• 13:17there was a sense of like,
• 13:18"Okay, a flood happened,
• 13:19there's going to be a bunch of malaria,
• 13:20and it'll come back down,"
• 13:21But didn't. Just kept going up and up and up.
• 13:23In the time since that flood,
• 13:25you've had several other destabilizing events.
• 13:282015, as you might be aware, was this mega El Nino,
• 13:32with global effects.
• 13:34You also had dengue and Zika,
• 13:36and particularly with the Zika scare
• 13:37coming through this region at that time,
• 13:40which really stressed health systems.
• 13:42And so, one thing that we're trying to do now is say,
• 13:45"Okay, in this context of intermingled climatic effects,
• 13:49social effects, epidemiological effects,
• 13:52what exactly is going on here?"
• 13:54And this is critical, because, you know, 10 years ago,
• 13:57if I were giving this talk 10 years ago,
• 13:58we'd be talking about elimination of malaria in the Amazon.
• 14:01We are not talking about that right now.
• 14:02We're talking about trying to control
• 14:03what seems to be an increase...
• 14:05Though you don't see it on this graph,
• 14:07because Peru seems to settle down a bit,
• 14:09not just an increase, but really,
• 14:11maybe a significant continuing increase of malaria
• 14:15in the region.
• 14:18Okay, so let me jump into the physical geography
• 14:22and hydrometeorology of the problem.
• 14:27So, let me start off with a little bit about the vectors.
• 14:29So as I will attempt to stress throughout this talk,
• 14:33when we talk about the influence
• 14:35of environment and hydrometeorology,
• 14:36we're not just talking about mosquitoes, okay?
• 14:40Mosquitoes are a big part of it.
• 14:42So, that's why I start off with them,
• 14:43but we always want to be thinking about mosquitoes.
• 14:45You want to talk about the pathogen,
• 14:47and we also want to talk about human behavior.
• 14:50Nevertheless, the influence of land cover
• 14:53in hydrometeorology in particular
• 14:55on an anopheles mosquitoes is going to be
• 14:57a big part of our story,
• 14:59so I want to make sure you're familiar
• 15:00with what's going on in the Amazon.
• 15:02So, the red here is showing anopheles darlingi.
• 15:05That is the dominant malaria
• 15:09competent vector in the Amazon.
• 15:14There are a whole bunch of others,
• 15:15a great diversity of anopheles mosquitoes here,
• 15:18but the darlingi is going to be the number one.
• 15:23And if we zoom in a little bit,
• 15:24so just a little box there,
• 15:25around this portion of the Western Amazon,
• 15:28centered on the Laredo district of Peru,
• 15:31which is kind of the Northern Amazonian district in Peru,
• 15:35you can go and study this there,
• 15:38because a lot of really good work has been done
• 15:41by some of the members of the team
• 15:42that were on my title slide,
• 15:43and people who preceded them or partnered with them
• 15:46in this area doing really strong work on mosquito surveys,
• 15:51or collecting or doing species typing.
• 15:54And this happened along various areas in the region.
• 15:58And I don't know how well this is showing up on your screen,
• 15:59but that red inset there is a Landsat satellite snapshot
• 16:04of the area.
• 16:05And you might see red dots, yellow dots, green dots.
• 16:09Those are all showing collection sites
• 16:11where breeding habitats and mosquito species types
• 16:13were collected at larval and adult stages.
• 16:16And they were organized along transportation corridors,
• 16:19these surveys.
• 16:20And so the red dots are along a highway
• 16:22that connects Iquitos to Nauta, a town to the south.
• 16:26The yellow dots connect Iquitos to Mozan up in the north.
• 16:30And then the green dots are going along various rivers
• 16:33that are used as transportation corridors.
• 16:36Let me just zoom in on that a little bit,
• 16:39so you get a sense of this region.
• 16:41So here, this is just kind of a true color satellite image
• 16:45of what I showed in the previous slides.
• 16:47You see the Amazon river flowing south to north here
• 16:50through the region.
• 16:51That urbanized area that you see
• 16:53along the banks of this meander is Iquitos.
• 16:58Iquitos is famously the largest city in the world
• 17:01that you can not reach by road.
• 17:03You either have to come in on the river
• 17:04or you have to fly in.
• 17:05The rivers are the dominant transportation networks,
• 17:07but we have these roads I showed before.
• 17:10There's one to the north that kind of cuts off
• 17:11here, going to Mozan,
• 17:13but this highway here, the Iquitos to Nauta highway
• 17:17is kind of the biggest road in the area.
• 17:18And you see that herringbone deforestation
• 17:21coming along that road.
• 17:24And so, what we have here are mosquito collections
• 17:27in an area of land use contrasts,
• 17:31including the pristine forest
• 17:33and breeding into areas of significant agricultural activity
• 17:36and urban activity.
• 17:39And so, we can then use our satellite images
• 17:41to classify the different types of cover we see here,
• 17:45and these range from different water types.
• 17:46We always want distinguish between clear water
• 17:48and silky water in the Amazon.
• 17:49They're very different ecologies.
• 17:52And then different kinds of Amazon basin land cover type,
• 17:57including the anthropic types,
• 17:59such as disturbed vegetation and bare ground,
• 18:02and roads and buildings,
• 18:03and the natural vegetation types,
• 18:04including different types of forest.
• 18:07Okay.
• 18:07And so when we analyze these together,
• 18:10the land cover information with the mosquito information,
• 18:14you find some interesting patterns.
• 18:17And what I have here are all anopheles species.
• 18:20Okay, I didn't bother spelling out all
• 18:22of the species names, because they're long
• 18:23and it doesn't matter too much.
• 18:25But what this box plot is intended to demonstrate
• 18:28is that, as your forest area decreases, okay,
• 18:32as you go down on the Y axis into the negative area here,
• 18:36you will see decrease.
• 18:38You will see different relationships with different species.
• 18:42Okay, and when you have a...
• 18:46Sorry, I apologize. Let me step back.
• 18:48The Y axis here is the association. Okay?
• 18:51And so you see negative associations
• 18:53between forest area and some species,
• 18:55and positive associations between forest area
• 18:59and other species.
• 19:01Okay.
• 19:02And so, what's interesting about this is that you say,
• 19:04"Okay, there's going to be changing species assemblages,
• 19:06as land cover shifts from natural forest
• 19:11to more cleared area."
• 19:13But it's somewhat systematic,
• 19:14in that the species here over to the left
• 19:18are the malaria competent species.
• 19:20You'll see anopheles darlingi here on the far left.
• 19:23And so, that's a dominant vector and all of these others
• 19:25are vectors, also.
• 19:27These are not, okay?
• 19:29So it so happens that as you clear forest,
• 19:32you might not actually see an increase
• 19:33in the total number of anopheles mosquitoes.
• 19:35You often will see a decrease in the total number
• 19:37of mosquitoes of all species,
• 19:39but you'll see an increase in the prevalence
• 19:42and absolute number of darlingi, of your vector species.
• 19:45And in fact, it's even quantified.
• 19:47Here's some data we had.
• 19:48We found that for every 1% increase in clear land area,
• 19:51you have close to a 4% increase in the odds
• 19:53of finding anopheles darlingi at a collection site.
• 19:57So we have here is human wildlife interface
• 20:01causing more mosquito human interactions.
• 20:05And also, the anthropic disturbances of the landscape
• 20:08increasing the proportion of your competent vectors.
• 20:13So this is a recipe for increased malaria transmission.
• 20:16So this is a fairly detailed study
• 20:17that we could only do in places where we had
• 20:19really detailed entomological collections.
• 20:23We don't have that everywhere,
• 20:25but at least from the satellite perspective,
• 20:26we can take this kind of last
• 20:28and done at high resolution and zoom out of it.
• 20:31And so as we try to look across all of the Laredo states,
• 20:36this shows Laredo state of Peru,
• 20:37and this analysis has now been extended to include
• 20:40the Amazonian portions of Ecuador,
• 20:42as well as parts of Colombia and Brazil.
• 20:46We can make use of satellite data.
• 20:49And here I'm showing the MODIS satellite data.
• 20:51If you're not familiar with MODIS,
• 20:52it's a NASA-supported mission has been up
• 20:54for about 20 years now.
• 20:56And unlike the previous images that I showed,
• 20:58which is a Landsat higher resolution, 30 meter resolution,
• 21:01but you only get snapshots every once in awhile,
• 21:04MODIS is giving you 250 to 500 meter resolution,
• 21:08but it's giving you daily images.
• 21:09And these really cloudy areas that's important, right?
• 21:11So you need to catch when you can
• 21:13a view through the clouds.
• 21:15And that allows us to use phenology.
• 21:16That is the seasonality of the vegetation
• 21:19to do a more detailed classification of land cover types.
• 21:22And it says on the left, just a classification using MODIS.
• 21:25We can then, because the satellite's been up for 20 years,
• 21:28look at change in these forest types over time.
• 21:31All of that can go into our malaria risk analyses.
• 21:35And on the right, what I'm showing you is a card
• 21:37that I did not develop,
• 21:38that NatureServe developed,
• 21:40which used a combination of satellite data
• 21:42and measurements on the ground to come up
• 21:44with ecological systems,
• 21:46that we view as potentially relevant to malaria.
• 21:49In particular, the red areas on this map
• 21:52are areas that are forested,
• 21:54that are flooded by what they called black water.
• 21:56So those tannic waters of the Amazon.
• 21:58And then in the light green,
• 22:00you'll see other areas that are flooded
• 22:01by what they're calling white or clear water.
• 22:03Might have sediment in it, but it's not tannic, okay?
• 22:05So again, different water quality, different ecology.
• 22:11And so, what I've taken here is land use,
• 22:13look at really high resolution land use,
• 22:15to understand the scale of distribution.
• 22:17Used a different satellite assets in order to zoom out
• 22:19and say, "What can we say at scale about land use
• 22:23and vegetation types?"
• 22:25And also, thanks to the NatureServe analysis,
• 22:28link that somehow to hydrology, right?
• 22:32Because now we're talking about ecological zones
• 22:34that are defined, in part, by their flooding regime,
• 22:37which is a key consideration in the Amazon, right?
• 22:40There's a lot of forest
• 22:41that's different from other forests,
• 22:43and much of that has to do with these flooding regimes.
• 22:46So this brings hydrometeorology into the picture, right?
• 22:48And so, how does hydrometeorology matter?
• 22:51As I mentioned, it's going to affect the vector, right?
• 22:53We're concerned about breeding sites,
• 22:54survivability of different life stages,
• 22:56the life cycle, speed of the life cycle of the mosquito,
• 23:00dispersion of mosquitoes,
• 23:01influenced by winds and temperature.
• 23:04And so, wind, temperature and certainly precipitation
• 23:08and moisture conditions in the soil and surface puddles
• 23:10are going to be a big deal.
• 23:11We also know the plasmodium has temperature sensitivities,
• 23:15and that the vector's competence transmit the plasmodium
• 23:18is a function of temperature.
• 23:21On top of that, you've got human behavior.
• 23:23And so migratory labor in particular,
• 23:25logging in this area is very sensitive to the river height,
• 23:29because you need the rivers to be a certain height
• 23:31in order to float the logs downstream.
• 23:33And so that will have an influence.
• 23:34And then of course, agricultural activities
• 23:36will be sensitive to the seasonality of hydrometeorology,
• 23:42as well as the inter-annual variability.
• 23:45When you get interventions,
• 23:45you also have an issue that anyone
• 23:48who's worked in malaria knows, which is,
• 23:49"Will people use bed nets?"
• 23:51And when it gets really hot, very often,
• 23:53it gets harder to comfortably use a bed net.
• 23:58So, how are we going to do hydrometeorology?
• 24:00So there are a lot of different ways you can do this.
• 24:03The system that my group uses,
• 24:05and kind of one of our major contributions
• 24:07to this malaria problem is called
• 24:09the land data assimilation system.
• 24:11So that probably doesn't get discussed too much
• 24:13at schools of public health, which is appropriate.
• 24:15So let me give you a little background,
• 24:17because this is an area where any of you
• 24:19potentially working on various climate environment
• 24:23influence on disease,
• 24:25but really any host of public health issues
• 24:28might be able to make use of such a system,
• 24:31collaboratively or on your own,
• 24:34to really bring environmental data in, in a powerful way.
• 24:37So what an LDAS does is it merges observations
• 24:39with numerical models,
• 24:41in order to get your best possible estimates
• 24:42of what's going on with the land surface
• 24:44and the lower atmosphere than your surface meteorology.
• 24:47Why do you do this?
• 24:48You do this because satellite observations
• 24:51are amazingly powerful tools, but they're snapshots
• 24:54of single variables.
• 24:56And so, if you want a comprehensive view
• 24:57of what's happening with all the potential
• 24:59variables of interest, you kind of want a model, right?
• 25:02You want something to give you spatially
• 25:03and temporally complete and consistent representation.
• 25:10But those models don't necessarily represent reality,
• 25:12particularly in data limited environments, like the Amazon.
• 25:16And so what you do with an LDAS is you basically
• 25:19pick at the best of both worlds to the extent possible.
• 25:22You have an advanced, physically based model
• 25:24that is trying to simulate what's going on
• 25:26with your weather and with your hydrology.
• 25:28And then you've got satellite observations
• 25:30that inform that model and kind of keep it realistic.
• 25:34And so, in schematic form,
• 25:36what you have is a bunch of landscape information,
• 25:38such as the land cover analyses I've just shown you,
• 25:41often satellite-derived.
• 25:43You have meteorological data,
• 25:44which is also often from satellites,
• 25:46or from other weather analysis systems.
• 25:50Those all drive a numerical model,
• 25:53which is then going to produce estimates
• 25:55of energy balance and hydrology, okay?
• 25:57So that'll get you, you know,
• 25:58the temperature, radiation, wind, moisture conditions
• 26:02you care about.
• 26:03As you run this model forward, you assimilate observations.
• 26:07And so you can update observations.
• 26:08So for example, information about soil moisture variability.
• 26:12Graded estimates come from satellite
• 26:14can be brought into the numerical model
• 26:16to update the model's estimate of soil moisture.
• 26:19And so, you end up with a system.
• 26:21This should be obvious,
• 26:22because we're using updated observations.
• 26:24This isn't like a future projection model, right?
• 26:27The model itself might be able to,
• 26:28but the LDAS system is retrospective,
• 26:31up through real-time monitoring,
• 26:33where you're bringing in these update observations,
• 26:34because the observations you can only have
• 26:37after we've taken the observation.
• 26:39Okay?
• 26:40And so these LDS systems are in a lot of places, you know?
• 26:45It's related, first of all, to weather forecast.
• 26:47Weather forecasts use LDAS, as well as assimilation
• 26:50of atmospheric variables.
• 26:51So those are used all the time.
• 26:53We also use these LDAS in a lot of the work we do,
• 26:56for example, on agricultural monitoring
• 26:58in the United States,
• 27:00climate assessment reports are very often include LDAS,
• 27:04like the National Climate Assessment of the United States.
• 27:06Work we do with the Famine Early Warning System in Africa.
• 27:09These LDAS are known to be pretty useful ways
• 27:10to get information.
• 27:12And so some of them have outputs that are available,
• 27:16that you can just get,
• 27:17because there's already someone running it.
• 27:18If you're interested in that,
• 27:20please contact me and I'll try to put you in touch.
• 27:22And then sometimes we run them ourselves
• 27:24to optimize them for a region we have here.
• 27:27There's a couple more minutes on this,
• 27:30just so you understand the basic principles here.
• 27:34One of the most important starting points
• 27:36is satellite-derived rainfall.
• 27:37We're using a couple of products here.
• 27:39I'm not going to bother with the acronyms.
• 27:40They don't matter.
• 27:41They are, in case anyone attending today
• 27:42is from the satellite world and is interested
• 27:44in what we're using, okay?
• 27:45So CHIRPS and GPM-IMERG.
• 27:49We then use that MODIS satellite that I already described,
• 27:52get our land cover and vegetation characteristics.
• 27:54And this cartoon here is showing you our model.
• 27:57It's called the Noah MultiParameterization
• 27:59Land Surface Model.
• 28:01And what it's doing is it's simulating
• 28:02multiple layers of the soil,
• 28:04different vegetation types, shallow groundwater.
• 28:08We also work into it a downscaling routine
• 28:10to get better surface meteorological estimates.
• 28:13It doesn't simulate the atmosphere,
• 28:14but it can help to downscale atmospheric conditions.
• 28:19And it also does snow, which actually does matter to us
• 28:23because we want to get the runoff coming out of the Andes,
• 28:24but it doesn't matter locally in the Amazon, obviously.
• 28:29So, that's all kind of for the local energy
• 28:32and water balance solution.
• 28:33We use Noah MP.
• 28:33We then couple it to a river routing model called HyMAP.
• 28:38And HyMAP, the hydrological modeling and analysis program
• 28:42that's what that stands for,
• 28:43allows us to model things like the flood plain,
• 28:45and that's, as you can imagine,
• 28:46really critical when you're talking
• 28:47about mosquito habitats.
• 28:49So we get the river heights.
• 28:50We also get the river width,
• 28:51and the area of flooded river boundary at any given time.
• 29:00We run this at five kilometer, gritty resolution.
• 29:02Five kilometers by five kilometers, or 25 square kilometer.
• 29:05And then around Iquitos,
• 29:06that city that has the largest population center.
• 29:09We nest into one kilometer
• 29:11for some higher resolution analysis.
• 29:15As we run the model forward,
• 29:16we can take advantage of these assimilation capabilities,
• 29:18and we run multiple simulations for different purposes.
• 29:21Sometimes we might be assimilating satellite-derived
• 29:24estimates of soil moisture, or leaf area index,
• 29:26or water storage, terrestrial water sources,
• 29:28meaning all the water stored in the soil column
• 29:30and groundwater.
• 29:31These are all observables at different resolutions
• 29:34from space using different civilian space missions.
• 29:39And those will all help to improve the performance
• 29:41of our model.
• 29:42And then you can get an output like what I'm showing
• 29:43on the right-hand side of the screen here,
• 29:44which is just a standardized anomaly in soil moisture,
• 29:47showing a period where, in our area of interest,
• 29:49for example, there were some drought going on
• 29:52in the Northwestern Amazon,
• 29:53as shown by a negative standardized anomaly
• 29:55in soil moisture, as captured by our system.
• 30:00I'll come back to this in a moment,
• 30:01but this particular snapshot is an interesting example,
• 30:05and that's showing what might be considered
• 30:08the classic El Nino pattern, okay?
• 30:10So it's an old snapshot. This one's from 1998.
• 30:13I've accidentally cut the date off of it.
• 30:15There's the monthly anomaly from a month in 1998.
• 30:19And what you're seeing here is the 1997, '98 El Nino
• 30:22bringing catastrophic flooding to the coast
• 30:25of Peru and Ecuador, and drought to the Amazon basin.
• 30:28Okay, I'll return to that in a moment,
• 30:31but that's kind of a classic El Nino pattern in the region.
• 30:36And so, here's just a quick animation
• 30:37to show what you're getting through time.
• 30:39I'm showing monthly up what's here.
• 30:40In fact, we get, you know,
• 30:42hourly outputs from the system that we can then extract
• 30:47for different geographies to perform our malaria analysis.
• 30:51Information on things like your air temperature anomaly,
• 30:53your rainfall, your soil moisture anomaly,
• 30:55your runoff, your river height, et cetera, et cetera.
• 30:57Okay, and so this is all the information
• 30:59that we're going to be bringing in,
• 31:01combining with the land cover and ecological information,
• 31:04to try to get this environmentally informed malaria analysis
• 31:09and early warning systems set up.
• 31:12So, one thing that you might be wondering is,
• 31:15"Okay, I just mentioned this was a data scarce area, right?"
• 31:19And these are outputs of some system
• 31:21that's combining satellite data with its uncertainties,
• 31:23and a model with its own uncertainties.
• 31:26How good is it, right? And can you trust it?
• 31:29And the answer is that in any study you do,
• 31:32where you want to make use of this
• 31:33kind of environmental data,
• 31:35you want to make sure that either you or someone else
• 31:37has done a good, clean analysis of how well
• 31:40that system performs in your region
• 31:43and season of interest, okay?
• 31:45You don't want to just take this off the shelf and say,
• 31:47"Oh, I know this going, going to be fine where I am."
• 31:49And so we've done some analysis.
• 31:53I'm not going to make you sit through
• 31:55our whole analysis kind of thing that we spend our days,
• 31:58nights and weekends doing, right?
• 31:59Make sure the systems work well
• 32:00and trying to fine tune them.
• 32:03But we have some data here that Cristina Recalde,
• 32:05a PhD student working with me has from Ecuador,
• 32:09and some data from Peru, looking at things like,
• 32:11"Okay, how well do we do in observations in blue,
• 32:15versus our model simulation on rainfall?"
• 32:18And there are good and bad things
• 32:19if you stare long enough at this chart,
• 32:21like, yeah, we're in about the magnitude
• 32:23of rainfall is not bad.
• 32:24The seasonality is pretty good most places,
• 32:26but then you'll find there's some wet and dry bias
• 32:28in different places that we're investigating.
• 32:31Similarly, you can then look at the soil moisture.
• 32:33Soil moisture is harder, because rainfall,
• 32:35there actually are rainfall observations.
• 32:37Not many, but there are some, right?
• 32:40Soil moisture, there's like basically
• 32:42no in-situ observations in a consistent way
• 32:44in the study area,
• 32:46and so we have to use satellite data to compare it to.
• 32:48So here, we're comparing this observation in gray,
• 32:50which is really a satellite observation,
• 32:52with our model simulation.
• 32:54And again, seeing some good, some bad.
• 32:57Here, we really do have to question the fidelity
• 32:59of both the observation and the model,
• 33:01since the observation is satellite-derived.
• 33:03At least it gives us a sense.
• 33:04Do we have a consensus across our different estimates,
• 33:07as to what's going on here?
• 33:10And this is tricky, right?
• 33:11Because getting soil moisture right in a complex hydrology
• 33:13like the Amazon is no trivial task.
• 33:16So this is a scenario where we spend a lot of our effort.
• 33:21Last point I want to make on this physical hydrology
• 33:24hydrometeorology before finally getting
• 33:26just the natural malaria results:
• 33:29it's really important,
• 33:31whenever you're doing a study like this, right,
• 33:34to distinguish between,
• 33:36when I say that there's hydrometeorological variability,
• 33:39am I talking about geographic variability?
• 33:41You know, wet versus dry places.
• 33:43Am I talking about seasonal variability, right?
• 33:46A wet season versus the dry season, for example.
• 33:48Or am I talking about something
• 33:50like inter-annual variability?
• 33:52Like, "Oh, we had a drought year,
• 33:53or we had a year with more flooding."
• 33:56And that's really important, you know,
• 33:58first and foremost, to understand process, right?
• 34:01You want to know that you get a statistical result
• 34:03that comes out of throwing some environmental variables
• 34:06into your model.
• 34:08They're significant. What is it that you're seeing?
• 34:10Right?
• 34:12And also, is what you're seeing a proxy for something else?
• 34:16Right?
• 34:17If you classically see like,
• 34:19"Oh, there's a wet versus dry season response,"
• 34:20or a warm versus cold season response,
• 34:23and when I look at my cases of malaria,
• 34:26is that because temperature's affecting malaria,
• 34:28or is it because there's a seasonal cycle in temperature,
• 34:31and seasonality for some other reason
• 34:33is affecting the malaria, and I'm calling it temperature?
• 34:36Okay.
• 34:37And so, you want to be clear on whether you're looking
• 34:41at the geography, the season,
• 34:42or the inter-annual variability.
• 34:44And this is on my mind a lot these days,
• 34:48both because I do a lot of this work.
• 34:51And as I know Kai appreciates and probably others
• 34:52in the audience as well,
• 34:54there's a lot of conflation of these things
• 34:56in the COVID-19 literature,
• 34:58with different claims or attempts to claim
• 35:00environmental sensitivities.
• 35:02Some really good work, right?
• 35:04But also a lot of these kind of naive, I would say,
• 35:06studies that came out showing correlations
• 35:08or associations that were simply showing a seasonality,
• 35:11or, you know, a coincidence of two patterns.
• 35:13The whole correlation versus causation problem,
• 35:16that I think part of the problem there
• 35:19was a misunderstanding or there's a mis-framing
• 35:23of what kind of climatic variability we're talking about.
• 35:28Okay, got off that soap box.
• 35:31And simply say for that third thing,
• 35:32all I've shown you here is seasonality
• 35:35and spatial variability.
• 35:37I haven't shown you inter-annual variability.
• 35:38I want to comment a little bit on that in this region,
• 35:41because anyone who's worked on malaria in the Amazon
• 35:44or other malaria zones probably are aware
• 35:46of a lot of studies, good studies, right?
• 35:50That have associated malaria
• 35:52with various large scale climate modes.
• 35:55Certainly these drivers of variability, okay?
• 35:59And so the big one is El Nino.
• 36:01The El Nino Southern oscillation, okay?
• 36:04But there are many others.
• 36:05It's an alphabet soup that I won't get into.
• 36:08El Nino, in this part of the region.
• 36:11One might well expect an El Nino effect here, right?
• 36:13It's called El Nino because of the effects it had,
• 36:16you know, was first characterized in the coast of Peru,
• 36:18and what it does to the sardine fisheries
• 36:20off the coast of Peru.
• 36:21And so, this is kind of like the home of El Nino, right?
• 36:24And so, we certainly expect an El Nino effect.
• 36:26And as I showed a few slides ago,
• 36:28a classic pattern would be high rainfall on the coast,
• 36:31drought in the Amazon,
• 36:32for dynamical reasons that I won't get into.
• 36:37It's not that simple or that predictive
• 36:41as a simple univariate association
• 36:44in this part of the Amazon, at least.
• 36:47There's some other parts of the Amazon
• 36:48that respond a little bit more reliably,
• 36:50but I'll tell you, it's always a little complicated.
• 36:53But here, just taking it again from Cristina's work here,
• 36:57looking at El Ninos and La Ninas the past 20 years.
• 37:01And if it's red, it means you've got drought,
• 37:02or drier conditions.
• 37:04If it's blue, it means you have wet anomalies.
• 37:06And again, during El Nino, we should be seeing red
• 37:08in the Amazon.
• 37:09And here, you know, we got our Laredo state.
• 37:12Sorry, it was just Ecuador and Peru I'm showing you.
• 37:15So we've got this kind of, here's your Northern Amazon
• 37:18portion of our study region.
• 37:21And what you're seeing is that, yeah, during some El Ninos,
• 37:24you do see that drought pattern, okay?
• 37:26But you also see it in this La Nina,
• 37:29and then there are some El Ninos
• 37:31where you don't see it at all,
• 37:32and in fact, that big monster El Nino that hit in 2015
• 37:36and had effects globally, it was wet
• 37:39in our part of the world,
• 37:41when you might've thought it was supposed to be dry.
• 37:44And so, there are some complications here, okay?
• 37:49All I can say that one could use, and so El Nino,
• 37:54La Nina oscillations effectively, statistically,
• 37:57in a forecast in here,
• 37:59if you accounted for enough other variables.
• 38:01I'm highlighting the fact that it's not enough
• 38:04of a predictor of rainfall in its own right, okay?
• 38:06But combined with other factors,
• 38:08you can probably get some scale.
• 38:10But we decided to take a different approach,
• 38:12which is, rather than using these kinds of teleconnections,
• 38:15these like remote connections to El Nino directly
• 38:17in our model,
• 38:18we run a dynamically based forecast.
• 38:21And so what we're doing there is, again,
• 38:24this one's a little detail for those who might be
• 38:26working at this interface of climate and health.
• 38:29We run what we call subseasonal to seasonal forecast.
• 38:32You know, a few weeks out to...
• 38:34Well, you can go to nine months.
• 38:35We're really only going up to three months right now,
• 38:36for this application.
• 38:39And what you do is you take what I already showed you
• 38:41in the LDAS, the satellite landscape analysis,
• 38:43run it through a land data simulation system.
• 38:46That provides initial conditions,
• 38:49from which you generate an ensemble.
• 38:50So your seasonal forecasts are
• 38:52these probabilistic ensembles, just like weather forecasts.
• 38:55And these are these global atmospheric models
• 38:57that we run forward.
• 38:59We run them forward using initial conditions
• 39:01of the hydrology locally, and the ecology locally.
• 39:05We downscale the meteorology
• 39:07from those global forecast systems
• 39:09using some algorithms that, again, I won't get into,
• 39:12but happy to follow up with anyone doing this kind of work.
• 39:15And then, we put that into hydrologic work.
• 39:17As we run it through the same modeling system,
• 39:19it's no longer data simulation
• 39:21because we don't have observations.
• 39:22We run that system forward.
• 39:24Okay. So why do all of this?
• 39:27Because it pushes your forecast time horizon out.
• 39:34If I, as the climate guy in the team,
• 39:36give Bill and Mark, the epidemiology guys on the team,
• 39:39a monitoring system that is operationally saying
• 39:41what the moisture is right now,
• 39:43they can forecast malaria because it's a time lag, right?
• 39:46So they'll get a pretty good forecast,
• 39:48because it takes time for the signal I'm sending them
• 39:50to propagate through the ecology, and the human systems.
• 39:54But if I can give them a forecast of what it's going to
• 39:56be like in two months, that gives them, you know,
• 39:58eight weeks more lead time,
• 40:00and you can make a different set of decisions,
• 40:01given an extra two months, right?
• 40:03So it's all about this uncertainty time horizon
• 40:06trade-off year.
• 40:07The more we push out for a greater time horizon,
• 40:09the greater our certainty,
• 40:10but also potentially the greater power
• 40:13of the decision-making
• 40:14that that kind of system can empower.
• 40:18So, how did these forecasts look?
• 40:22I'm not going to make you sit through
• 40:23a whole forecast scale analysis,
• 40:24but just want to make one point here.
• 40:26If you just focus, let's say, on correlation here,
• 40:29for the sake of time,
• 40:31if there's hashing,
• 40:32it means a statistically significant scale.
• 40:34And what you see here is that looking at something
• 40:36like soil moisture, we get really good forecasts
• 40:40for one month, and then it begins to degrade,
• 40:43particularly degrading these wet areas.
• 40:46You've maintained some forecast scale out in the dry areas,
• 40:48because there's so much memory, right?
• 40:49If it's not raining much,
• 40:50most of the initial conditions that matter.
• 40:53But as you go out,
• 40:54the result here we might say is that
• 40:56we can really do a nice job of getting you
• 40:58an extra four weeks, right, on the system.
• 41:00If you want eight weeks or 12 weeks,
• 41:02and you know, we're not going to be contributing
• 41:04that much stuff in the forecast.
• 41:05And so it's important both to have the capability,
• 41:07and to understand the limitations of the capability.
• 41:10All right.
• 41:11So we do all those analyses.
• 41:14And then, this is not my work.
• 41:16This is work that Bill led.
• 41:17He took all of this ecological and hydrological analysis,
• 41:23and did an objective regionalization,
• 41:25did principal components analysis on the variability.
• 41:28End up with these three different factors
• 41:29that are loaded by different properties of the system,
• 41:33and counting for about, you know, human systems,
• 41:35as well as land use and hydrometeorological conditions.
• 41:40And from that, derived these seven
• 41:42socioenvironmental regions.
• 41:45And the principle here is that these regions
• 41:48are reasonably homogeneous and regionally distinct
• 41:51from each other,
• 41:51with respect to human and environmental factors.
• 41:55And also, as it happens,
• 41:57this was not necessarily integrated to that,
• 41:58but because you've included the human systems
• 42:00in the analysis, most of the travel stays within the region.
• 42:05And you typically have similar vector species
• 42:08within a region.
• 42:09Okay, and similar livelihoods.
• 42:11So, what we then say we're not going to develop
• 42:13one malaria risk model.
• 42:16And again, this is now, we're seeing Laredo regions,
• 42:18so this part of Peru.
• 42:19We're going to develop a system that has customized models,
• 42:23based on socioenvironmental regions.
• 42:27So, in the remaining time that I have, which isn't much,
• 42:29I know, so I'll touch on these lightly,
• 42:32but these are just examples of how we can
• 42:34pull this all together, all right?
• 42:36And so the first thing,
• 42:38kind of the motivation for this whole presentation,
• 42:40this whole project is forecast, right?
• 42:44And so, using these socioenvironmental regions,
• 42:47then aggregate malaria data,
• 42:49which we have about 300 health posts contributing data,
• 42:52passive surveillance.
• 42:54They get aggregated to a socioenvironmental region.
• 42:56And then we try to predict whether there's an outbreak,
• 42:58based on the Ministry of Health's definition
• 43:01of what an outbreak is, which is, you know,
• 43:04exceeding a certain threshold,
• 43:05in terms of case number per population.
• 43:09Again, this work led out of Duke by Bill,
• 43:11and he uses observed components model
• 43:13as a statistical method,
• 43:15and was seeking to get a time horizon of four to 12 weeks.
• 43:20And again, because it's customized by region,
• 43:22what you'll find is that the model
• 43:24has different variable importance
• 43:27and is structured differently for the different models.
• 43:28So region one, which includes Iquitos,
• 43:31so it's kind of like our most urban area,
• 43:34we can describe that in terms of the characteristics
• 43:36of the socioecological region.
• 43:38And then we can say, "Okay, what explanatory variables
• 43:41from our environmental suite end up being significant?"
• 43:44It turns out to be soil moisture.
• 43:46We can then look at a region like region three,
• 43:48kind of really out in the forest,
• 43:49very low population density.
• 43:51It has a different description.
• 43:52It's going to have different statistical characteristics
• 43:54to this unobserved components model.
• 43:56And in this case, minimum temperature
• 43:58came out of the more significant variable.
• 44:00Both of these variables, of course,
• 44:01if you look at the literature, are using malaria prediction.
• 44:04So they're both plausible, they're possible pathways,
• 44:06but different ones came out as more predictive
• 44:08in these different regions.
• 44:11Okay? So then we run the system.
• 44:15We have to run the system starting four weeks
• 44:17before the present.
• 44:18Why?
• 44:19Because it takes about four weeks
• 44:22for surveillance to come in.
• 44:23Here's the percent of health post reporting
• 44:26of malaria data.
• 44:27As you can see, this is time, this is the present.
• 44:31At the present, you have fewer than 20% reporting.
• 44:33If you go back four weeks,
• 44:34you have close to 100% reporting,
• 44:36which means that you have a good...
• 44:37You know, previous cases are important predictor
• 44:39of future cases.
• 44:41So the forecast includes a four week forecast of the past.
• 44:45And then, we want to go out to eight or 12 weeks
• 44:47in the future.
• 44:49We have this moving outbreak threshold,
• 44:51because it varies seasonally and by location,
• 44:53what MINSA, the health ministry decides
• 44:55is the right threshold to declare an outbreak.
• 44:57And then we might have an observation,
• 44:59and a competence interval around that observation.
• 45:03Just to give you an example of performance,
• 45:042016 was the first year we really tried this.
• 45:07So this isn't just a systematic analysis,
• 45:09just showing you the kinds of things you look at.
• 45:11True positives, false negatives, false positives,
• 45:13true negative.
• 45:15For an outbreak in any of these eco regions,
• 45:18looking at eco region one and three here,
• 45:20over the different forecast time horizons,
• 45:22our sensitivity and our specificity.
• 45:25In a nutshell, we do really well in eco region one.
• 45:28Fades a little in specificity as we get out
• 45:30to 12 week time horizon, still pretty good.
• 45:33eco region three, we do not do that well, okay?
• 45:37And so again, small sample one year,
• 45:39but just our first test was showing us
• 45:41that we're going to get different performance
• 45:42in different eco regions.
• 45:46Okay.
• 45:47And so, that's all at the eco region level.
• 45:51I'm not going to get to too many more results
• 45:52at that level right now,
• 45:54but rather say that to be decision relevant,
• 45:56we have to go down to the district level.
• 45:58So, the lines here on this map are separating the districts.
• 46:02Okay.
• 46:03And so the colors of the eco regions
• 46:04aligns with the district.
• 46:05We really want to be at a district level.
• 46:07And so for this, again, won't get to the details right now,
• 46:10but Mark Janko implemented this hierarchical
• 46:13Bayesian spatio-temporal logistic model,
• 46:16where you basically have your district outbreak probability
• 46:20being a function of the probability of an outbreak
• 46:23in the eco region that contains the district,
• 46:25and some district-specific properties.
• 46:29When Mark downscaled and looked at some of these analyses
• 46:31and then did an evaluation over a retrospective period,
• 46:36these are the kinds of sensitivities and specificity
• 46:38we're getting for different districts
• 46:40within each eco region.
• 46:41Again, just showing you eco region one and three here
• 46:43as examples.
• 46:44And you'll see that again, pretty high variability.
• 46:47So we were doing well in eco region one at eco region level,
• 46:50but you'll see that, for example,
• 46:51in the district of Fernando Loris,
• 46:53there were some pretty significant errors
• 46:55in this retrospective period,
• 46:58and different kinds of errors in different places.
• 47:00So also for us to look at,
• 47:02in eco region three, kind of uniformly doing worse
• 47:05in general, than eco region one.
• 47:08So why is that? Why are we doing poorly in region three?
• 47:10Multiple reasons.
• 47:11One thing I want to emphasize is that eco region three
• 47:16was very much located kind of up in this area.
• 47:18So first of all, malaria cases are generally low there
• 47:20in total, because it's such a sparsely populated area.
• 47:23But it's also a border area.
• 47:24It's a border area that is transected
• 47:27by trans boundary rivers.
• 47:28The trans boundary rivers are the transportation
• 47:31in the region.
• 47:33And so what we find is that our model fits most poorly here
• 47:36in eco region three and another eco region
• 47:39dominated by trans boundary river.
• 47:41Doesn't do well in places along the rivers. Okay?
• 47:45And so that's one big weakness in the model
• 47:48that we're working on.
• 47:51And oops, the slides got reversed.
• 47:54And I just want to point out that we are looking at,
• 47:57and we had a paper recently, led by students.
• 47:59And so this is students from Duke, Johns Hopkins,
• 48:02Ecuador and Peru, who took the initiative
• 48:04to really lead an analysis of this cross-border spillover.
• 48:08And that's something we're looking at now.
• 48:11Okay.
• 48:13So, that's where the forecast system is.
• 48:15We brought it in 2019.
• 48:17We did some operational forecasts for the Health Ministry.
• 48:20Was all looking good.
• 48:21Then there's political change and COVID,
• 48:23so we're a little bit on hold right now,
• 48:24but we've got a system that we've proved
• 48:26we can use operationally.
• 48:27We continue to try to improve the performance.
• 48:30Policy evaluation. Okay.
• 48:33So I'm going to give one example
• 48:35of policy analysis we've done.
• 48:38That was PAMAFRO, which was this project for malaria control
• 48:41on the Andean border areas, active 2006 to 2010 or 11,
• 48:45depending on how you counted.
• 48:47They did four kinds of things.
• 48:48Long-lasting insecticidal nets,
• 48:51better rapid diagnostic tests, and other monitoring tools,
• 48:56case management, with antimalarial drugs and training,
• 49:00and environmental management for vector control.
• 49:02So doing these four kinds of things.
• 49:04And it kind of worked, right?
• 49:06So this is by vivax and falciparum in Laredo.
• 49:09And it sure looks like over the PAMAFRO period,
• 49:11the case counts were going down, down, down,
• 49:13approaching eradication, which was the goal of the program.
• 49:17Then stops suddenly in 2011, cases start coming back up.
• 49:21And what we can do is we can leverage that district model
• 49:24that Mark Janko developed, right?
• 49:27Not only using it for forecasts, but then saying,
• 49:29"Well, let's include in that model structure
• 49:31the different interventions, especially with PAMAFRO."
• 49:33Because we know at district level and with monthly timing,
• 49:37what kind of interventions were done where.
• 49:39Let's integrate that to a model and then do
• 49:41an interrupted time series analysis,
• 49:44and see what those interventions actually accomplished
• 49:47on the background of climate variability,
• 49:50and all the other variables in our model.
• 49:52So kind of an environmentally controlled analysis
• 49:55of the effectiveness of the intervention.
• 49:58Mark's found is that, well, you can kind of quantify this.
• 50:02So the blue line here in the top left,
• 50:04top is vivax, bottom is falciparum.
• 50:07Blue lines are the model, dots are the observation.
• 50:11On the left, we have the PAMAFRO period.
• 50:13And we see that our model,
• 50:15if you don't tell it about the intervention,
• 50:16systematically overestimates the cases in this period,
• 50:19for both vivax and falciparum.
• 50:21In the post PAMAFRO period, starting in 2011,
• 50:25quite the opposite.
• 50:26Our model has cases down here.
• 50:28The observed cases were much higher.
• 50:32And so, take those together and come up with estimates
• 50:36that about 150,000 cases were averted by PAMAFRO.
• 50:38That was the amount of malaria averted thanks to PAMAFRO,
• 50:42and had you continued it for another five years,
• 50:45you would've averted another 150,000,
• 50:47not to mention the long-lasting impact
• 50:49of driving cases that low, right?
• 50:52And so here we have an analysis of both the effectiveness
• 50:55and the cost of removing a program
• 50:58without a good continuity plan.
• 51:01And then you can zoom in, because again,
• 51:03we have this district level information
• 51:04on each kind of intervention.
• 51:06I see I'm running out of time,
• 51:07so I won't spend too much time walking through these maps,
• 51:09but green shows incidence ratio less than one.
• 51:13And so we can look district by district
• 51:14and say, "Okay, for falciparum and vivax,
• 51:20for each of the four intervention types,
• 51:22environmental management, bed nets, et cetera,
• 51:25in which districts do we see the most effect
• 51:27when we add or remove this from our interpretive
• 51:29time series analysis?"
• 51:31And there's some interesting patterns that appear
• 51:32that we're in conversation with some of our partners about
• 51:36to figure out what might be effective in the future.
• 51:40One of the cool thing just mentioned
• 51:41that you can do with this
• 51:42is try to figure out how much malaria and dengue there is
• 51:46right now in this area, because we have no idea.
• 51:49If you look at what happened in 2020 with surveillance,
• 51:52I mean the health system basically shut down.
• 51:54And so, it looks like it was a great year
• 51:55for malaria control, but of course it wasn't.
• 51:59So we can then use this same modeling approach
• 52:02to try to estimate how many cases there really were
• 52:04in the year, 2020 and 2021.
• 52:06And as you can see, we estimate that there were
• 52:08at least three times as many cases.
• 52:13Okay.
• 52:14Last point I want to make here is that
• 52:17I've showed you some malaria modeling cases
• 52:18that are process-informed,
• 52:20but at their heart, statistical, right?
• 52:22These are empirical analyses.
• 52:24And looking at intervention scenarios,
• 52:26we are also looking at explicit simulation of behavior,
• 52:31okay, to get these coupled natural human systems right.
• 52:34And the way that we are doing that,
• 52:36led by Francisco Pizzitutti,
• 52:38is with agent-based modeling.
• 52:40And this is a kind of Coolidge based model Francisco built,
• 52:42in that it has agents that are mosquitoes, humans,
• 52:46and plasmodium, okay?
• 52:47So. you have all of these are agents interacting.
• 52:50And here is just an example of one of the villages
• 52:52where he's applied this,
• 52:53where you can have different households,
• 52:56and all these agents are interacting
• 52:58and influenced by the environment.
• 53:00In that here, we see different kinds of breeding habitats
• 53:03influenced by seasonal flooding,
• 53:05with information from our environmental analysis system,
• 53:08changing the hydrology.
• 53:09And then you've got the cases happening in this household,
• 53:11each of which is also experiencing
• 53:13its own environmental conditions, okay?
• 53:16You can then run scenarios of control.
• 53:18For example, vector control strategies,
• 53:21one thing we like to look at.
• 53:22And so we're looking at here
• 53:23at one of these environmental control applications,
• 53:25and saying, "Well, what if you do larval habitat control
• 53:28around a certain buffer radius,
• 53:30around each household, right?"
• 53:32How well do you do at 50 meters, 100 meters,
• 53:34150 meters, 200 meters,
• 53:36when you talk about malaria incidents?
• 53:37Total vivax falciparum.
• 53:39And the idea here is that,
• 53:40by understanding this agent based model
• 53:43movement patterns, right?
• 53:45And the sensitivities of the different agent types,
• 53:49we can get a sense, say,
• 53:50"Well, really you want to probably get out
• 53:52while you take your pick,
• 53:53but I would say at least 150 meters
• 53:55might be considered very effective.
• 53:56Anything beyond 200 is unnecessary."
• 53:59All right.
• 54:00And so this is parametrized for one set of villages.
• 54:03It's very data intensive, but nevertheless,
• 54:05I think it indicates a powerful way to,
• 54:07you know, use your environmental information
• 54:09in a different manner, not as an empirical predictor,
• 54:12but as a variable within a model
• 54:16in which different agents are responding
• 54:17according to decision rules to this variability.
• 54:24You can also use the same tool, and Francisco has,
• 54:26to look at the importance of mobility, right?
• 54:28So that's something people talk a lot about
• 54:29in the past couple of years, right?
• 54:30How much mobility influences disease transmission.
• 54:33It's an old story from malaria.
• 54:34What you'll see here is if you look
• 54:35at your observed black line here
• 54:37of the average monthly malaria incidents
• 54:39along the Napo river,
• 54:42first thing you know, is that,
• 54:43"Well, okay, if I run this model with no asymptomatic cases
• 54:46considered in travel,"
• 54:47you assume that no asymptomatic people are traveling,
• 54:50you way underestimate the incidence rate.
• 54:52So we know there's a lot of asymptomatic activity going on.
• 54:56And then we can say,
• 54:57"Okay, as the percent of traveling workers increase,
• 55:00we would expect the incidence rate to increase."
• 55:03And we're right about the right order of magnitude.
• 55:04And it looks like some of this movement
• 55:06really does need to be accounted for,
• 55:07to understand the incidence rates
• 55:10with significant implications, again,
• 55:12or how you would do monitoring and control in the region.
• 55:16So, ran a little longer than I wanted to. Sorry.
• 55:19That's what happens when you let professors talk.
• 55:22But just a few of the next steps here.
• 55:24I break them into four categories.
• 55:26We're really working on the application here.
• 55:28As I noted, there's been a lot of political turnover
• 55:31in Peru for those who know the region,
• 55:32which has hampered our ability to operationalize a forecast.
• 55:35So now, we're starting to train and transfer
• 55:37to some universities and research institutions
• 55:40in the region, rather than straight to the government,
• 55:42to be able to spare stability.
• 55:45We're just having our first meeting this week
• 55:47on an effort to expand to include Columbia and Brazil.
• 55:50So it's a big up-scaling of the effort.
• 55:53And we're also seeing,
• 55:54can we transfer this to an area in central America,
• 55:57working with the Clinton Health Access Initiative, sorry.
• 56:01Flipped the letters.
• 56:04On Central America, where the case counts are low
• 56:06and therefore the ecology and the environmental sensitivity
• 56:09of the system shifts.
• 56:11It seems to cross a threshold.
• 56:12So we want to see how the approach works there.
• 56:15And last, but certainly not least,
• 56:17through these combined methods, but again,
• 56:19all trying to leverage the power of the different fields
• 56:21to understand malaria sensitivities.
• 56:24How can we continue to explain these coupled
• 56:25natural human mechanisms, which,
• 56:28despite the fact that we've known about these relationships
• 56:31since ancient times,
• 56:32we continue to struggle to understand
• 56:35in a predictive manner today.
• 56:37So, thank you again for the opportunity to talk.
• 56:39I realize I didn't leave too much time for questions,
• 56:41but maybe we have time for a couple.
• 56:51<v Kai>Thank you, Ben, for the great talk.</v>
• 56:52So, we actually have a class right after this seminar,
• 56:55so I think we only have time for one question,
• 56:59and the students have already read the papers
• 57:02that you mentioned published in your page.
• 57:06So, any of you want to ask a question directly?
• 57:11(indistinct)
• 57:13Okay, so let me ask you this question.
• 57:15So Ben, you gave wonderful talk on the importance
• 57:21of value, time and migrating,
• 57:24the importance of having the data,
• 57:26and then from the very state of the art
• 57:29subseasonal to seasonal forecast.
• 57:32The students when they read the paper, they have question
• 57:35regarding (indistinct) also COVID-19 related.
• 57:38So, did you see how to apply this malaria focus system?
• 57:45The application to COVID-19 control focus system?
• 57:52<v ->Yeah. Interesting point.</v>
• 57:54So, I'm going to answer in a very general way.
• 57:58They're obviously very different diseases, right?
• 58:00We're talking about a vector-based tropical disease
• 58:02versus a pandemic virus with a lot of airborne transmission.
• 58:07But I would say that the general challenge
• 58:10of bringing these different data sets together
• 58:12is really critical.
• 58:13And we can do cross-learning across diseases,
• 58:16because one thing we've really struggled with in COVID
• 58:19is to bring all the information together
• 58:20in systematic databases for responsible analysis.
• 58:24And we were able to leverage some of the things
• 58:26we've done with malaria and other tropical diseases,
• 58:29to build COVID information databases, to support research.
• 58:32And I know that Kai did his own work
• 58:33to pull his own database together.
• 58:35So moving forward,
• 58:36how can we use all of these diseases
• 58:37to inform those kinds of data structures,
• 58:39I think would be...
• 58:41And cross-learning approaches will be the way to go.
• 58:43I wouldn't necessarily endorse any single thing
• 58:45that I did here on malaria as the answer for COVID-19 model.
• 58:48They're too different.
• 58:49But if you can really focus on that kind of
• 58:53informed integration, I think there's a lot to be learned.
• 58:57<v Kai>Thank you so much, Ben.</v>
• 58:58And thank you, guys, for coming today,
• 59:00and thank you for our online audience.
• 59:03And just kind of reminder that today's lecture
• 59:07is recorded and will be available online,
• 59:10on our (indistinct) websites, so you can check that.
• 59:15Want to sincerely thank you, Ben,
• 59:17for giving this incredible talk.
• 59:21<v Benjamin>Great, thank you.</v>